Skip to content

Robotics · 12 November 2026 · 3 min read

Runway releases Praxis-1 weights under an Apache 2.0 licence

Runway said on 11 November that Praxis-1, a 27-billion-parameter model that predicts video and robot actions together, is available with open weights. On Nvidia’s RoboLab-120 it lists a 46.2 percent success rate across 1,200 episodes. The weights, inference code and fine-tuning code are under Apache 2.0, including commercial use.

Two robot arms over a white table, one holding a blue cloth beside wooden trays.
A photograph published with coverage of Praxis-1, showing two arms at a table. It is a demonstration setup. It is not the RoboLab-120 score.

Runway said on 11 November that Praxis-1 is available with open weights under the Apache 2.0 licence, including for commercial use. The company describes it as a 27-billion-parameter mixture-of-experts diffusion transformer that predicts video and actions together. It said launch partners Noble Machines, Standard Bots and Ultra had already run it on their own hardware: a legged humanoid, a six-axis industrial arm and a dual-arm warehouse robot. Runway says the model runs on Nvidia hardware from data-centre GPUs, including H100 and B200, and workstation GPUs, including RTX 5090 and RTX PRO 6000, down to Jetson Thor on the robot. The release includes weights, inference code and fine-tuning code, on GitHub, on Hugging Face with LeRobot support, and in vLLM-Omni, with hosted inference on Reactor.

Runway’s table for Nvidia’s RoboLab-120, dated as of November 2026, lists a 46.2 percent success rate for Praxis-1 across 1,200 episodes. The same table lists 42.9 percent for FLUX 3 Action, 36.8 percent for Cosmos3 Nano and 28.0 percent for π0.5. Those rates are Runway’s account of the leaderboard.